Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPT-4 API General Availability

openai.com

Post reply on HN
Print view View original

531–540 of 562 posts

Re: GPT-4 API General Availability

#531

Post by staticman2 »

logicchains wrote: post #499
staticman2 wrote: post #200

Earlier quoted context omitted.

There was a guy in the news who asked an AI to tell him it was a good idea to commit suicide, then he killed himself. Even on this forum I've seen AI enthusiasts claiming AI will be the best psychologist, best school teacher, etc.

That was Eliza, an AI so old that it's included in stock Emacs, not an LLM. It's propaganda, not news.

The chatbot app was called "Eliza" but it's not the Eliza you are thinking of.

https://news.ycombinator.com/item?id=35402777

https://www.businessinsider.com/widow-accuses-ai-chatbot-rea...

Re: GPT-4 API General Availability

#532

Post by jabradoodle »

kaba0 wrote: post #521
jabradoodle wrote: post #506

Earlier quoted context omitted.

> Life can’t be “won” by intelligence Humans being the dominant life form on Earth may suggest otherwise. > I honestly don’t see what changes here — super-human intelligence has limited benefits as it scales. Would you suddenly have more power in life, were you twice as smart? If so, we would have math professors as world leaders. Intelligent humans by definition do not have super human intelligence.

We know that this amount of intelligence was a huge evolutionary advantage. That tells us nothing whether being twice as smart would continue to give better results. But arguably the advantages of intelligence are diminishing, otherwise we would have much smarter people in more powerful positions. Also, a big tongue in cheek but someone like John von Neumann definitely had superhuman intelligence.

On a serious note though a person with an IQ of 150 can't clone themselves 10k times.

They also tend to have some level of autonomy in not following the orders of idiots and psychopaths.

Re: GPT-4 API General Availability

#533

Post by superkuh »

LeoPanthera wrote: post #406
superkuh wrote: post #84

Yikes. They're actually killing off text-davinci-003. RIP to the most capable remaining model and RIP to all text completion style freedom. Now it's censored/aligned chat or instruct models with arbitrary input metaphor limits for everything. gpt3.5-turbo is terrible in comparison. This will end my usage of openai for most things. I doubt my $5-$10 API payments per month will matter. This just lights more of a fire u…

Everyone who complains about being "censored" never gives examples.

Anyone who doesn't has never actually toyed with LLM and received, "As an AI language model I can't" in response to an innocuous request to, say, write a limerick about a politician or list the reasons why "username2 is stupid".

But mostly it has to do with the fact that LLM do what they've seen. And if they've been fine-tuned to not respond to some classes of things they'll misapply that to lots of other things. That's why most people go for the "uncensored" fine tuning datasets for the llamas even for completely sfw use cases.

Re: GPT-4 API General Availability

#534

Post by ben_w »

kaba0 wrote: post #527
ben_w wrote: post #525

Earlier quoted context omitted.

> But arguably the advantages of intelligence are diminishing, otherwise we would have much smarter people in more powerful positions. Smart people get what they want more often than less smart people. This can include positions of power, but not always — leadership decisions come with the cost of being responsible for things going wrong, so people who have a sense of responsibility (or empathy for those who suffer f…

That doesn’t contradict what I said. There is definitely a huge benefit to an IQ 110 over 70. But there is not that big a jump between 110 and 150, let alone even further.

Really? You don't see a contradiction in me saying: "get what they want" != "get leadership position"?

A smart AI that also doesn't want power is, if I understand his fears right, something Yudkowsky would be 80% fine with; power-seeking is one of the reasons to expect a sufficiently smart AI that's been given a badly phrased goal to take over.

I don't think anyone has yet got a way to even score AI on power-seeking, let alone measure them, let alone engineer it, but hopefully something like that will come out of the super-alignment research position OpenAI also just announced.

I would be surprised if the average IQ of major leaders is less than 120, and anything over 130 is in the "we didn't get a big enough sample side to validate the test" region. I'm somewhere in the latter region, and power over others doesn't motivate me at all, if anything it seems like manipulation and that repulses me.

I didn't think of this previously, but I should've also mentioned there are biological fitness constraints that stop our heads getting bigger even if the IQ itself would be otherwise helpful, and our brains are unusually high power draws… but that's by biological standards, it's only 20 watts, which even personal computers can easily surpass.

Re: GPT-4 API General Availability

#535

Post by ineedasername »

ben_w wrote: post #474
ineedasername wrote: post #355

Earlier quoted context omitted.

> embarrassingly parallel I don’t see why such a thing should be embarrassing. Or, at least no more so than being acute or obtuse. Just as long as nothing is askew.

"Embarrassingly parallel" is a term of art: https://en.wikipedia.org/wiki/Embarrassingly_parallel

Yes I was anthropomorphising it back into the realm of human emotion, wherein the angles at which one’s lines run need not be a source of emotional distress. Excepting perhaps the innate sadness of two parallel lines destined to ever be at each other’s sides but still never to meet across the infinite plane.

Re: GPT-4 API General Availability

#536

Post by tensor »

rvcdbn wrote: post #505
tensor wrote: post #242

Earlier quoted context omitted.

It doesn't have to be copyrightable to be intellectual property.

I’m going to go out on a limb here and assume that you’re making this statement because it feels like they should have some intellectual property rights in this case. Independently of whether that feeling corresponds to legal reality (the original question) I would also encourage you to question the source of this feeling. I believe it is rooted in an ideology where information is restricted as property by default. T…

It has nothing to do with any sort of feeling. Perhaps you should check your own mental state.

It is the same as any confidential data. Logs, readings from sensors, etc etc. If it's confidential and given to a 3rd party through a contract that doesn't mean that it's suddenly not confidential data for the rest of the world, even if the 3rd party leaks it.

And if you really have a lawyer trying to tell you that some, at best, extreme grey area, is fine to build a business on, I think you should find a new lawyer.

Re: GPT-4 API General Availability

#537

Post by tensor »

PostOnce wrote: post #253
tensor wrote: post #233

Earlier quoted context omitted.

A reminder that llama isn't legal for the vast majority of use cases. Unless you signed their contract and then you can use it only for research purposes.

OpenLLaMA is though. https://github.com/openlm-research/open_llama All of these are surmountable problems. We can beat OpenAI. We can drain their moat.

Absolutely, 100% agree. I just wouldn't touch the original LLaMA weights. There are many amazing open source models being built that should be used instead.

Re: GPT-4 API General Availability

#538

Post by PostOnce »

wokwokwok wrote: post #419
PostOnce wrote: post #372

Earlier quoted context omitted.

Different quantizations can give you a big speedup if you've had "depressingly slow" issues. Even the slowest ones (that fit in RAM) will run at basically interactive speed, not instant, but also not "email speed". I have a laptop with a 2018 CPU and I'm working with them just fine. Text generation style instead of chat style is another avenue that makes the feedback time not so annoying for a developer. at 100ms/tok…

Whaaaaat, how are you getting 100ms per token on an 5 year old potato without a graphics card? Like, not vaguely hand wavey stuff, specifically, what model and what inference code? I get nothing like that performance for the 7B models, forget the larger models, using llama.cpp on a pc without an nvidia GPU.

Here is a short test of a 7B 4bit model on an intel 8350U laptop with no AMD/Nvidia GPU.

On that laptop CPU from 2017, using a copy of llama.cpp I compiled 2 days ago (just "make", no special options, no BLAS, etc):

  ./main -m models/WizardLM-7B-uncensored.ggmlv3.q4_0.bin -n 128 -s 99 -p "A short test for Hacker News:"

  llama_print_timings:      sample time =    19.12 ms /    36 runs   (    0.53 ms per token,  1882.65 tokens per second)
  llama_print_timings: prompt eval time =   886.82 ms /     9 tokens (   98.54 ms per token,    10.15 tokens per second)
  llama_print_timings:        eval time =  5507.31 ms /    35 runs   (  157.35 ms per token,     6.36 tokens per second)
and a second run:

  ./main -m models/WizardLM-7B-uncensored.ggmlv3.q4_0.bin -n 128 -s 99 -p "Sherlock Holmes favorite dinner was "

  llama_print_timings:      sample time =    54.37 ms /   102 runs   (    0.53 ms per token,  1875.93 tokens per second)
  llama_print_timings: prompt eval time =   876.94 ms /     9 tokens (   97.44 ms per token,    10.26 tokens per second)
  llama_print_timings:        eval time = 16057.95 ms /   101 runs   (  158.99 ms per token,     6.29 tokens per second)
at 158ms per token, if we guess a word is 2.5 tokens, then that's 151 words per minute, much faster than most people can type. On a $250 laptop. Isn't the future neat?

the code I was running: https://github.com/ggerganov/llama.cpp

and the model: https://huggingface.co/TheBloke/WizardLM-7B-uncensored-GGML

There are other models that may perform better, I'm going to be doing a lot of screwing around with OpenLLaMA this weekend.

Post reply on HN